Force Alignment with original text

[TechMaster]

I generate a Vietnamese song with Suno. I want to generate a subtitle file that highlights each word spoken in the song.
I use WhisperX to transcribe this song

whisperx song.wav --language vi --model large-v2 --highlight_words True -f json

I have 2 questions:
1- Time stamp of full sentences are correct but time stamp of each word is incorrect. How should I fix this problem? Which parameter should I add in command line to make it work?

2- Some words in transcribed JSON file are incorrect. I have original lyrics of the song. How can I input original lyrics to WhisperX to improve accuracy of the json file?